Development and Validation of AI Algorithms for Dental Radiography Interpretation

Abstract

This study developed and validated AI algorithms for for orthopantomogram (OPTG) interpretaNon, comparing accuracy with radiologists. An experimental study design was used in a dental seRng for this research. A sample of 138 orthopantomogram (OPTG) were divided into two groups (AI and Human) using straNfied random sampling. The performance of AI and human-AI was evaluated using confusion matrices, sensiNvity, specificity, accuracy, and error analysis conducted in R Studio. The study analyzed 138 individuals (mean age 44.59 years) with a balanced gender distribuNon (50% male, 50% female) and varying severity levels (mild, moderate, severe). Diagnoses included caries, fractures (roots of teeth), and other abnormaliNes such as periodontal disease, traumaNc lesions, and neoplasms (benign and malignant). The AI model showed beYer performance than the human control model in all the important markers, such as sensiNvity (86.84 vs 79.41), specificity (90.32 vs 88.57), and accuracy (88.4 vs 84.1). The AI showed beYer results than the above parameters in the ROC curve (AUC 0.886 vs 0.84). Compared to the interpretaNons made by humans, AI has proven to be more accurate in its diagnosNcs and has a higher sensiNvity and specificity rate.